Shivanshu Shukla

I build machine learning and generative AI projects spanning NLP, computer vision, and transformer-based language models. My work includes a Hinglish sentiment classifier, Quora duplicate-question detection, a character-level GPT built from scratch, and plant disease detection.

For Hinglish sentiment analysis, I combined BERT-based tokenization, TF-IDF features, Bidirectional LSTM attention pooling, and ensemble models, then deployed inference through a FastAPI REST endpoint. I also built a Quora similarity pipeline across 400K+ question pairs using 20+ NLP features and cross-validated Logistic Regression, XGBoost, and Random Forest models.

I implemented a decoder-only Transformer in PyTorch with causal self-attention, positional embeddings, custom tokenization, checkpointing, and controlled text sampling. Over 5,000 iterations, I reduced validation loss from 4.29 to 1.66.

As DSA & Machine Learning Coordinator at CP-BYTE Technical Club, KIET, I've mentored 50+ students and led AI/ML workshops for 25+ students. I'm AWS-certified across four credentials, have solved 1,300+ coding problems, and earned first place at the Innotech Hackathon.